Although we got bits and pieces of supersecondary structure, nothing
we came up with looked like a complete model.

We felt that the conserved histidines, H174, H176, and H187 ought to
cluster, perhaps at a metal-binding site, but they did not cluster
naturally in any of our models.  When we forced clustering, we lost
predicted secondary structur.


This is try16-opt2, the undertaker-generated model that scores best
with cost functions that require clustering the histidines (excluding
models derived from robetta).